|
I'm looking at building a new machine, and want to get on the GPU passthrough bandwagon for gaming. Is there any hardware to avoid, and anything to actively go for?
|
#
?
Aug 17, 2017 22:05
|
|
|
|
| # ? Apr 26, 2024 19:43 |
|
|
Horse Clocks posted:I'm looking at building a new machine, and want to get on the GPU passthrough bandwagon for gaming. Ryzen still has problems with Passthrough if you are using KVM. (You can't enable NPT correctly). It works on Xen, but then you have to work around the nVidia "bug" https://github.com/sk1080/nvidia-kvm-patcher Passthrough is significantly easier if you have two different GPU chipsets (like two of nvidia, amd, intel).
|
|
#
?
Aug 17, 2017 22:23
|
|
|
Horse Clocks posted:I'm looking at building a new machine, and want to get on the GPU passthrough bandwagon for gaming. You probably want an Intel CPU and an AMD GPU for the easiest GPU passthrough experience.
|
|
#
?
Aug 17, 2017 22:29
|
|
|
SoftNum posted:Ryzen still has problems with Passthrough if you are using KVM. (You can't enable NPT correctly). It works on Xen, but then you have to work around the nVidia "bug" https://github.com/sk1080/nvidia-kvm-patcher I would like to play with virtualizing hardware as well so I'm in a similar boat, if anyone's got a guide they followed that they know works with recent drivers/whatever please link it. Or at least let me know whether I should be looking at ESXi or KVM or what, I've done plenty of playing around with VirtualBox inside my normal OS but I don't know the first thing about actual server-grade virtualization. Mostly I am curious about running Windows and FreeBSD virtualized in the same system so I could have native ZFS going without needing a separate server box. That doesn't necessarily imply that I need virtualized graphics hardware (could run HyperV inside Windows) but I'm also kinda curious about actual legit multi-seat gaming setups. Depending on the game I could probably run two entirely separate copies on my current CPU (5820K), and an extra GPU is no problem either way.
|
|
#
?
Aug 17, 2017 22:33
|
|
|
SoftNum posted:Ryzen still has problems with Passthrough if you are using KVM. (You can't enable NPT correctly). It works on Xen, but then you have to work around the nVidia "bug" https://github.com/sk1080/nvidia-kvm-patcher Twerk from Home posted:You probably want an Intel CPU and an AMD GPU for the easiest GPU passthrough experience. This is my fear. I'm pretty much building this machine because I have an excuse to use CUDA for work. Which kinda limits me to Nvidia GPUs. Might just bite the bullet, skip the cool factor, and dual-boot. I'm guessing I'd need to do a full power cycle anyway if I want to switch between work on a linux VM, and gaming on a windows VM.
|
|
#
?
Aug 17, 2017 22:58
|
|
|
There's a recent YouTube video by Level1Linux where he explains how to get the most out of Ryzen passthrough. I'm using Intel/NVIDIA but it's worth a look. The two boards he's using look good because they've plenty of segregated IOMMU groups. For anyone new to this it's worth pausing and rewinding because he goes pretty fast but gives a good overview of how it's set up.
|
|
#
?
Aug 17, 2017 23:10
|
|
|
I'll be doing the whole GPU passthrough gaming thing once Volta comes out, so at least I have the luxury of waiting so I can see if that kvm bug gets any traction or not. I'd rather do AMD but poor vm preformance isn't acceptable, and I'd really prefer kvm over xen. Here's the bug in case you want to follow it.
|
|
#
?
Aug 17, 2017 23:54
|
|
|
Horse Clocks posted:I assumed this was the case, and I have a AMD R7, and Geforce 720 kicking around, just for the occasion. nvidias aren't really that much worse than AMD. I think KVM hides the virt identifier by default so with an intel processor you should be fine. If you have 2 monitors (or a KVM switch) and 2 graphics cards, there's no need to reboot or anything. Passthrough is really fine these days.
|
|
#
?
Aug 18, 2017 00:50
|
|
|
Desuwa posted:I'll be doing the whole GPU passthrough gaming thing once Volta comes out, so at least I have the luxury of waiting so I can see if that kvm bug gets any traction or not. I'd rather do AMD but poor vm preformance isn't acceptable, and I'd really prefer kvm over xen. There's a thread about this in vfio-users. Looks like it may actually be a bug in the qemu i440 firmware somewhere. Paul MaudDib posted:I would like to play with virtualizing hardware as well so I'm in a similar boat, if anyone's got a guide they followed that they know works with recent drivers/whatever please link it. Or at least let me know whether I should be looking at ESXi or KVM or what, I've done plenty of playing around with VirtualBox inside my normal OS but I don't know the first thing about actual server-grade virtualization. Get a system with iommu support (vt-d or whatever amd calls it), both chipset and CPU. Follow one of a million guides. You should use whatever you're comfortable with. ESXi isn't gonna be usable as a workstation, where Linux+kvm is, so that's a selling point for many. Passthrough isn't actually virtualized graphics -- it's native. But you don't need passthrough for disks either, unless you want to present the entire controller to zfs. For multi-seat, 3 monitors, 2 keyboards, 2 mice. Or just get 2 GPUs and let people steam stream it. SoftNum posted:nvidias aren't really that much worse than AMD. I think KVM hides the virt identifier by default so with an intel processor you should be fine. It doesn't. You also need to turn off Hyper-V enlightenments, because Nvidia.
|
|
#
?
Aug 18, 2017 03:42
|
|
|
I have a server with a RAID-0 array where the hardware RAID superblock got clobbered but the data is intact. Is there any way to reconstruct this using software RAID in Linux so I can move the data off?
|
|
#
?
Aug 18, 2017 17:15
|
|
|
Can I bake certain installations into a Linux iso? At his request, I've set a friend up with some cryptocurrency wallets. When I'm watching him use his PC it makes me internally cringe and think "I wouldn't do that.", "you should really be doing x instead of y first". I'd like to bake a VeraCrypt installation into an Ubuntu iso so that it's preinstalled when he runs it off a USB stick. Then he can access his wallet without me worrying that he's using a malware-ridden PC. Is the main solution to use a persistent installation of Ubuntu? Because I'd rather create something that runs fresh every time.
|
|
#
?
Aug 19, 2017 07:42
|
|
|
apropos man posted:At his request, I've set a friend up with some cryptocurrency wallets. When I'm watching him use his PC it makes me internally cringe and think "I wouldn't do that.", "you should really be doing x instead of y first".
|
|
#
?
Aug 19, 2017 08:00
|
|
|
apropos man posted:Can I bake certain installations into a Linux iso? make sure to epoxy his network ports too
|
|
#
?
Aug 19, 2017 08:57
|
|
|
My friend wants to run Kdenlive video editor at work, which is populated with Windows machines. He specifically wants to run Kdenlive and won't run any other video editor. Is there a way I could go about baking Kdenlive into a Linux live ISO for him? I asked him why he didn't just take a netbook to work with him and he said he was gonna buy one but he's wasted all his money on bitcoins and even though bitcoins are rubbish and a handful of retailers would accept them for a netbook he said "no. Bitcoins are for holding" because he's provably clinically insane.
|
|
#
?
Aug 19, 2017 12:47
|
|
|
apropos man posted:My friend wants to run Kdenlive video editor at work, which is populated with Windows machines.
|
|
#
?
Aug 19, 2017 13:09
|
|
|
apropos man posted:My friend wants to run Kdenlive video editor at work, which is populated with Windows machines. Do people use Linux for video editing, specifically Kdenlive? Like are there any high-profile companies using this?
|
|
#
?
Aug 19, 2017 15:42
|
|
|
mike12345 posted:Do people use Linux for video editing, specifically Kdenlive? Like are there any high-profile companies using this? Mostly high-end systems. Autodesk Flame, Foundry Nuke, Blackmagic DaVinci Resolve and Fusion, etc... Blackmagic even has a turnkey CentOS-based ISO you can install to get running quickly.
|
|
#
?
Aug 19, 2017 16:12
|
|
|
apropos man posted:My friend wants to run Kdenlive video editor at work, which is populated with Windows machines. This is doable, but it's probably easier to just generate a liveUSB that supports persistence, then install kdenlive normally on that. E.g. with OpenSUSE, you'd:
And it'll automatically turn the extra space on the USB stick into a persistence partition where the stuff you installed (and any configuration changes you make, etc) is saved.
|
|
#
?
Aug 19, 2017 16:18
|
|
|
there's instructions for this for practically any distro. Ubuntu here: https://help.ubuntu.com/community/LiveCDCustomization In general you: * Extract the iso FS * chroot to the ISO * Install packages, gently caress with settings, etc. * unchroot * prepare / burn / make ISO
|
|
#
?
Aug 19, 2017 16:26
|
|
|
apropos man posted:My friend wants to run Kdenlive video editor at work, which is populated with Windows machines. His employer will be ok with him loving about with his work machine ?
|
|
#
?
Aug 19, 2017 16:35
|
|
|
For all of these "can I bake this in", the answer is yes, trivially. Ubuntu has tooling for this (or a good doc page if you want to do it manually) RPM-based systems can do it all with a kickstart and livemedia-creator/lorax very easily. SuSE used to have a web-based tool that I can't recall the name of. Studio? It was really nice.
|
|
#
?
Aug 19, 2017 17:01
|
|
|
Cheers for the link to the Ubuntu bake/chroot instructions. Gonna have a gently caress around with that. To anyone questioning why someone's running a video editor from a live iso at work, if you look at the replies to this post, you'll see that it was never about Kdenlive. e: Using the Ubuntu manual page now. Would someone care to explain what squashfs does, and how it's incorporated into a distro? Is it used in most distros? apropos man fucked around with this message at 18:58 on Aug 19, 2017 |
|
#
?
Aug 19, 2017 17:22
|
|
|
apropos man posted:Cheers for the link to the Ubuntu bake/chroot instructions. Gonna have a gently caress around with that. Squashfs is simply a way of jamming everything together into as little space as possible. Think of it like a mountable tarball with more work done to erase zero/duplicate blocks. They're extremely common in live images, and some projects (oVirt node as an example, possibly something similar to coreos also) use them to ship a/b system updates. I'm the project lead for node, so I know a lot more about squashfs than I need to, and I'm happy to answer any questions
|
|
#
?
Aug 19, 2017 23:33
|
|
|
evol262 posted:Squashfs is simply a way of jamming everything together into as little space as possible. Think of it like a mountable tarball with more work done to erase zero/duplicate blocks. So a live squashfs would be extracted as a template of the standard Linux file structure if you choose to install to the hardrive (/etc,/opt,/usr,/var)? But once it's extracted to disk it obviously becomes the persistent, changeable backbone of the OS? And squashfs also extracted into memory each time you run a live image, so that you're getting the same file structure every time you boot a live image? Not sure if I understand what oVirt node is aiming to achieve. Is it like a lot of small OSes, each performing a single or a couple of instances of virtualisation? Like Qubes OS is supposed to be very modular for security, so you could achieve a similar effect of running lots of small, segregated but interacting environments with oVirt node?
|
|
#
?
Aug 20, 2017 08:31
|
|
|
apropos man posted:So a live squashfs would be extracted as a template of the standard Linux file structure if you choose to install to the hardrive (/etc,/opt,/usr,/var)? But once it's extracted to disk it obviously becomes the persistent, changeable backbone of the OS? If you loopback mount it, live images almost always have an ext filesystem inside the squashfs, and that gets mounted on / squashfs is used because the used space trying to shove a plain filesystem image onto a CD which contains a "real" environment instead of a very small installer (and anaconda also uses squashfs) would make the installer much too large to ship. Using Anaconda as an example, the CentOS boot ISO (~150mb) is built from a kickstart like a normal CentOS image, from RPMs. Compare the installed size of a minimal CentOS image+multipathd+blivet+anaconda+X+... We use this in oVirt node because we can ship a complete system (~560 packages) in a 500mb RPM. So upgrading from 7.3 to 7.4 (for example) can be done with a plain "yum update", which pulls an entirely new system (~1.9gb on disk) which is already configured, installs it, creates filesystems and bootloader entries, etc. It's very, very space efficient. It's also not extracted. The squashfs is loopback mounted, then the ext filesystem is mounted on / (for livecds), so not additional space, memory, or CPU is required. apropos man posted:And squashfs also extracted into memory each time you run a live image, so that you're getting the same file structure every time you boot a live image? apropos man posted:Not sure if I understand what oVirt node is aiming to achieve. Is it like a lot of small OSes, each performing a single or a couple of instances of virtualisation? Like Qubes OS is supposed to be very modular for security, so you could achieve a similar effect of running lots of small, segregated but interacting environments with oVirt node? Node and oVirt engine are part and parcel. It's broadly equivalent to ESXi, except it's a standard Linux system configured to work with oVirt/RHV. A/B updates are nice for a lot of customers so they can easily roll back, and there's only a security process for a single product instead of every package which could go into CentOS. It's more like an appliance. This is also built with a standard kickstart and some small scripts, so it gets reused as the base image for other RPM-based appliances (Foreman discovery image, gluster appliance, etc). Most Linux users have no need for it, the same use cases as ESXi apply.
|
|
#
?
Aug 20, 2017 13:47
|
|
|
apropos man posted:My friend wants to run Kdenlive video editor at work, which is populated with Windows machines.
|
|
#
?
Aug 20, 2017 22:01
|
|
|
Anybody sending JVM stats to Cloudwatch? I was thinking of using https://github.com/awslabs/collectd-cloudwatch but wasn't sure if anybody had some real world experience. It doesn't seem to be very well maintained so I'm a little weary of using this awslabs offering. What else are folks using?
|
|
#
?
Aug 22, 2017 01:11
|
|
|
I just spent an entire day trying to figure out why the test Spacewalk server I'm standing up refused to work. It turns out it's due to a bug reported more than 4.5 years ago which is still unfixed  How on earth is there an open source project where Oracle is the only officially supported DB option?http://htfdidt.blogspot.com/2013/12/spacewalk-repomdxml-not-found-for.html https://bugzilla.redhat.com/show_bug.cgi?id=906224
|
|
#
?
Aug 23, 2017 17:07
|
|
|
That's why my group still uses a home brew provisioning package I cobbled together from scratch like ten years ago. Every so often we review newer stuff like cobbler or spacewalk or most recently satellite/foreman and all of it dumps you into a burning hell of clicking on bad gui's and praying every time you think about updating the software to fix bugs. So we continue to use my garbage solution because it Just Works. 
|
|
#
?
Aug 23, 2017 18:05
|
|
|
Spacewalk/Satellite5 is basically dead. I've had nothing but good experiences with Foreman, though once you add candlepin and pulp (to make it the same as Satellite6), it turns into a huge "I want to be a multi-tenant solution even though you almost certainly don't actually want that" clusterfuck. If you do want that for whatever reason, Katello is the integrated upstream product. I don't know why it seems so difficult for the developers to make something which has reasonable defaults. If I'm deploying an RPM-based distro, stick me with the default templates for kickstarting/PXE instead of making me select them for every new distro type. Same for deb/preseed. I don't want to have to sift through 4 pages of jumpstart/whatever to get there. Similarly, let me say "I use puppet/ansible/salt" and use that instead of making me install plugins and go through a weird configuration dance. At least the dns/dhcp proxy and PXE stuff from Foreman works really well once you finally get it set up how you want it, and the integration with oVirt/VMware/openstack for provisioning is good.
|
|
#
?
Aug 23, 2017 18:44
|
|
|
What's the best way to force Apache on CentOS 7 to wait until after the PostgreSQL or network services have started? Should I be editing the systemd initscripts myself, or is there something else I should be doing?
|
|
#
?
Aug 25, 2017 00:33
|
|
|
The glib answer is that you just edit Apache's unit file so that there's a Wants=postgres.service and Requires=postgres.service. However there's a subtle difference between "started" and "ready to accept requests". The former just means the binary has been launched, but may still be in an initialization state where it hasn't opened any sockets. There are a few ways to deal with this: 1) Quick and dirty: Edit your systemd units so that Apache starts after Postgres (Wants=/Requires=, etc) but also has an "ExecStartPre=/bin/sleep 5" to delay starting Apache and give Postgres some breathing room. Fragile and not recommended. 2) If you are using Postgres 9.6 or above, it can use the systemd libraries (specifically sd_notify) to tell systemd that it's ready to begin serving requests. If Postgres's unit file has Type=notify, then systemd will wait for this notification before starting up dependent services. Just use the unit file given in the Postgres docs 3) If you can't upgrade Postgres to 9.6, you can use socket-activation in tandem with systemd-socket-proxyd, which is tool used to workaround the fact that not everything supports sd_notify yet.
|
|
#
?
Aug 25, 2017 01:09
|
|
|
You don't need to edit the unit file, just put a drop-in alongside it.quote:Along with a unit file foo.service, a "drop-in" directory foo.service.d/ may exist. All files with the suffix ".conf" from this directory will be parsed after the file itself is parsed. This is useful to alter or add configuration settings for a unit, without having to modify unit files. Each drop-in file must have appropriate section headers. Note that for instantiated units, this logic will first look for the instance ".d/" subdirectory and read its ".conf" files, followed by the template ".d/" subdirectory and the ".conf" files there. Also note that settings from the "[Install]" section are not honored in drop-in unit files, and have no effect.
|
|
#
?
Aug 25, 2017 01:36
|
|
|
Okay, I'm having this weird issue and thought it was because Apache was loading before the Postgres and network services were up, but putting network-online.target or Postgresql-9.6.service in After or Wants doesn't seem to fix it. So, here's the weird issue: When I reboot this server, Apache doesn't load the fcgid script it should be loading, and the webpage goes to the "Testing 123.." page as if I hadn't set up Apache before. If I restart the Apache service, the webserver works fine and I can get to my website. I've tried turning the logging directive up to debug, but the only error in that there's no index.html. This error goes away after the service is restarted. The only way I can get the webserver to load correctly at boot is to reinstall NetworkManager. I'm guessing this is due to NetworkManager-wait-online.service. Is there a way to get the features of that NetworkManager-wait-online.service without having the entire NetworkManager stack installed?
|
|
#
?
Aug 25, 2017 21:15
|
|
|
I've been out shopping for a Samsung 960 Evo today. I want to migrate the CentOS installation on my 850 Evo to the 960. The current OS is encrypted with luks FDE/LVM. I want everything to be the same on my new drive. The new one is 500GB and the old 850 is 250GB. Should I just unplug the 850 and pretend I'm installing CentOS on the 960, then when Anaconda has finished boot with both the drives in and rsync everything over to the new drive, as suggested a couple of pages ago? I don't mind setting up luks manually with cryptsetup and allocating encrypted LVM's but I'm not too keen on doing the boot and EFI partitions manually. Should I just dd the entire drive and somehow try and expand the main luks partition afterwards? I'm going out for a cycle ride to try and help me think this through. e: Had a poke around my host, and there's very little to copy over really. A couple of network shares, ZFS drivers needed and a couple of systemd timers. I'm just gonna do it the manual way and copy over the required units and config files. My VM's can be imported later. apropos man fucked around with this message at 18:59 on Aug 26, 2017 |
|
#
?
Aug 26, 2017 15:43
|
|
|
Isn't mount -a ONLY supposed to mount directories that aren't already mounted? /etc/fstab quote://ServerBox/SFTP/PC /backups cifs uid=pc,credentials=/root/.smbuser,vers=3.0 0 0 Run "mount -a" once, /etc/mtab (correctly) looks like this: quote://ServerBox/SFTP/PC /backups cifs rw,relatime,vers=3.0,sec=ntlmssp,cache=strict,username=PC,domain=ServerBox,uid=1001,forceuid,gid=0,noforcegid 0 0 I made a change to another location in /etc/fstab, and then ran "mount -a" to mount the second share. It also mounted the first share again. Thinking that was weird, I ran "mount -a" several more times. Now my /etc/mtab looks like this: quote://ServerBox/SFTP/PC /backups cifs rw,relatime,vers=3.0,sec=ntlmssp,cache=strict,username=PC,domain=ServerBox,uid=1001,forceuid,gid=0,noforcegid 0 0 It keeps mounting the same share to the same directory, over and over and over. If I keep running "mount -a", I get this: quote:mount error(5): Input/output error Basically, ALL the connections on the Windows server have been exhausted from Ubuntu remounting the same share over and over. How do I prevent "mount -a" from attempting to mount a directory that is already mounted?
|
|
#
?
Aug 26, 2017 22:19
|
|
|
Would Linux be able to rebuild a ZFS volume built from FreeBSD? I built one with NAS4Free but I'm toying with the idea of moving everything to a regular Linux box because I don't like FreeBSD (mostly because I'm used to Linux more I think) but I don't have anywhere to store the files safely if I have to rebuild the volume from scratch.
|
|
#
?
Aug 27, 2017 13:04
|
|
|
Is there a generic "task distribution" framework where workers run a loop of getting a task from a server and executing it? I'm specifically thinking of batching video encoding across multiple computers on a network but it's a pattern that comes up pretty frequently. Obviously this is overkill for the specific task at hand but I think the long-term solution is I should figure out how to expose my computers via SLURM.
|
|
#
?
Aug 27, 2017 13:48
|
|
|
Where does parted get its config information from? I just 'cloned' CentOS to my new drive from the old 850 by using rysnc on most of the root directory and when I run parted to check alignment of my new drive it still thinks I've booted from the 850: code:
apropos man fucked around with this message at 18:49 on Aug 27, 2017 |
|
#
?
Aug 27, 2017 18:30
|
|
|
|
| # ? Apr 26, 2024 19:43 |
|
|
Paul MaudDib posted:Is there a generic "task distribution" framework where workers run a loop of getting a task from a server and executing it? I'm specifically thinking of batching video encoding across multiple computers on a network but it's a pattern that comes up pretty frequently.
|
|
#
?
Aug 28, 2017 05:23
|
|